Receiving ROCm execution provider not supported in this build #1130
Comments
|
Hi! The ROCm support is currently very experimental. You can try pass |
No, when I build with --use_rocm, the build script only build cpu codes. |
Different with building onnxruntime-rocm, build onnxruntime-genai-rocm even didn't use hipify to convert cuda codes(.cu) to hip codes(.hip). And it didn't try to find Please, U need to fix it. |
Also, u need to change the version of Microsoft.ML.OnnxRuntime.Rocm from 1.19.2 to 1.21.0-dev-20241211-0046-64d8e25b in cmake/ortlib.cmake at line 39, because it's wrong. |
|
True! I updated it to 1.20.0 and it was able to pull the Rocm library. Maybe I need to update to the one you mentioned @Looong01 ! Do you have any suggestions on how can we make the hip compilation successful for getting the Rocm working? |
Compared to the build procedure of onnxruntime, the cmake files, U will know that it still needs a big effort to fill the CMake compile codes for ROCm. It's lack of a lot of codes. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Describe the bug
Hi team,
I am currently building OnnxRuntime-GenAI from source for AMD GPUs. With some tweaks to the main branch, I was able to successfully build the
wheeland install the python package. I was able to use the Model builder to convert LLMs to ROCm supported models. But when I try to run the inference, I am seeing the following error -onnxruntime_genai.onnxruntime_genai.OrtException: ROCm execution provider is not enabled in this build.I am sure there is no issue with onnxruntime dependency because I received the following supported EPs from it -
Please suggest how I can solve this problem.
To Reproduce
Steps to reproduce the behavior:
Expected behavior
I expect no errors and LLMs should output same as they do in Nvidia GPUs.
Desktop (please complete the following information):
The text was updated successfully, but these errors were encountered: